One of the more sophisticated arguments against the idea that the Affordable Care Act drove overdose rates higher is that the Medicaid expansion states were ones particularly hurt by the financial crisis and its aftermath, or in which there is a generalized social downturn unrelated to the expansion of health insurance. Perhaps liberal state governments benefited politically from a harder downturn, and so were there to approve the Medicaid expansion, but it was the downturn that drove overdose rates up, not the Affordable Care Act. While the matching methods I used in the last post should account for this possibility to some degree, it still seems possible that there is a more general complex of social problems besetting counties in the Upper Midwest and Northeast that are the true driver of increased overdose rates, rather than easier access to prescription drugs. After all, over the last year and a half, since Angus Deaton and Anne Case released their study of middle aged white mortality, there has been broadened discussion of a set of seemingly interrelated causes of death labeled “dying from despair” by the media: alcoholic liver failure, suicide, and overdoses, all of which are rising among working class and rural non-Hispanic whites. One of the first posts on this blog, in fact, was my own little analysis of what characteristics correlated with increased mortality in this group.
Were this hypothesis to be the case- that it is broader “despair,” whether economically or culturally mediated, that is the cause of increased overdose rates in the Medicaid expansion states, rather than increased insurance- than we would expect that the other “deaths of despair” would rise in tandem with overdose rates in counties in the Medicaid expansion states.
They do not. Alcohol-related deaths have a small and statistically-insignificant rise in Medicaid-expansion counties relative to Holdout counties, about a tenth as large as the overdose rise when baseline characteristics overdose rates or the variables from the 2010 American Community Survey (details discussed in last post) are included. Suicide rates have either a slight (and statistically insignificant) rise or a slight and statistically insignificant fall in Medicaid expansion counties, depending which control variables are included. The simple fact is that this is a drug overdose-specific change, which provides evidence that it is access to insurance rather than generalized economic or cultural despair that is the driving factor here.
Here is the OLS Sample for Alcohol deaths: OLSsamplealcohol
And the estimated impact:
And here is the OLS Sample for Suicides: OLSsamplesuicideExcel
For comparison, the estimated percent effect of Medicaid expansion on overdoses were around ten times as big as alcohol and twenty times as large (and in the opposite direction) as suicide:
Given the tiny size (and lack of statistical significance, though Saint Gelman may strike me down) of the alcohol and suicide estimates, I think it is safe to say that the divergence in overdose rates we are observing for overdoses is not mediated by more generalized social dysfunction, but is instead plausibly driven by increases in insurance rates caused by the Affordable Care Act.
Another way of looking at this is that we can predict the 2015 suicide and alcohol death rates extremely well just from the 2010 death rates alone, an R-squared of roughly 0.75 in comparison with less than half that for overdoses. Suicides and alcohol-related deaths, as Case and Deaton observed, have had worrying increases. But unlike drug-related overdoses, nothing has changed to upset the rank order of which states and counties have the highest suicide rates and highest alcohol-related death rates: most counties rose more-or-less in tandem.
For overdoses, on the other hand, something happened to make some counties spiral out of control, even as others have been able to constrain the rate of increase.
My guess is that that something was Obamacare.
I used essentially identical methods to derive these estimates to those in the previous post.Here’s a zip AlcoholSuicideDrugs2015 with combined files for all three analyses.
And here are the regression outputs, for alcohol:-
|(1) Average difference without controls||(2)Average difference, controlling for baseline alcohol-related death rates||(3)Average difference, controlling for baseline alcohol-related death rates and 2010 American Community Survey variables|
|VARIABLES||Rate 2015||Rate 2015||Rate 2015|
|2010 Alcohol-related Death Rate||1.260***||1.127***|
|Population in 2010 (100,000s)||-0.25|
|White income ($1000s)||0.0102|
|Median income ($1000s)||-0.0676|
|Percent White in 2010||0.0174|
|Rate of SSI Receipt in 2010||-0.0274|
Robust standard errors in parentheses. Standard Errors account for clustering at the State Level.
*** p<0.01, ** p<0.05, * p<0.1
And for suicide:
|(1)Average Differences without Controls||(2)Average differences controlling for 2010 Overdose Rates||(3)Average differences controlling for 2010 Overdose Rates and 2010 American Community Survey variables|
Robust standard errors in parentheses. Standard Errors are clustered at the State Level.
*** p<0.01, ** p<0.05, * p<0.1